Model Context Protocol (MCP)

The seenode Model Context Protocol (MCP) server lets AI coding agents deploy and operate your stack from the editor: static sites, web, worker, and private apps (from Git or a Docker image), managed databases, storage, domains, projects, registry credentials, and billing.

Server URL

https://mcp.seenode.com/mcp

Transport is Streamable HTTP. Authentication is OAuth 2.1 with PKCE against seenode. Do not put API keys or Authorization headers in MCP client config.

Connect to seenode’s MCP server

1
{
2
  "mcpServers": {
3
    "seenode": {
4
      "url": "https://mcp.seenode.com/mcp"
5
    }
6
  }
7
}

claude mcp add --scope user --transport http seenode https://mcp.seenode.com/mcp

1
{
2
  "servers": {
3
    "seenode": {
4
      "type": "http",
5
      "url": "https://mcp.seenode.com/mcp"
6
    }
7
  }
8
}

1
[mcp_servers.seenode]
2
url = "https://mcp.seenode.com/mcp"

After you connect

1. Sign in to a seenode account if you do not have one yet.
2. Complete the OAuth consent when your client opens the browser.
3. For Git deploys, connect GitHub or GitLab on seenode (the agent can call get_git_connections / connect_git_provider and give you a browser URL).
4. Smoke-test with a read-only prompt, for example:
   • “What is my current seenode team? Use get_current_team.”
   • “List my seenode web applications.”
   • “List my static sites.”
5. Only then run mutating flows (create app or database, set secrets, add domains, billing).

Enable human confirmation for mutating tools in your client when available.

Authentication

The MCP server uses OAuth so agents act as you on your seenode team, with browser consent, refresh tokens, and no long-lived secrets in plugin manifests.

REST API keys (Getting an API Key) are for the HTTP API and CI. They are not used in the recommended MCP client configuration.

What you can do

Agents can:

• Deploy a GitHub or GitLab repository as a static, web, worker, or private application
• Deploy a Docker / OCI image to web, worker, or private (with team registry credentials when the image is private)
• Create and link managed MySQL / PostgreSQL databases (without pasting passwords into chat)
• Inspect builds, wait on or cancel deployments, read logs and metrics
• Manage environment variables, custom domains (web and static), projects, storage volumes, and billing/credits

Typical full-stack recipe

1. Optional: create_project
2. create_database → poll get_database_state until ready
3. Git: inspect_repository → create_application. Image: resolve_registry_image (and create_registry_credential if private) → create_application with source_type='image'. Static SPA: create_application(application_type='static', build_command, publish_directory, …) — no database, run command, or port
4. link_database_to_application (skip for static)
5. create_deployment / wait_for_deployment

Check get_credit_balance before paid creates. After rotate_database_user_password, call link_database_to_application again with overwrite=true.

Example prompts

• “Deploy this repository to seenode as a static site.”
• “Deploy this repository as a web app, create Postgres, and link them.”
• “Deploy ghcr.io/org/app:1.2.3 as a private service.”
• “Why did my latest seenode deployment fail?”
• “What’s my seenode credit balance in USD?”
• “Add example.com to my seenode web app and show the DNS records.”

Money in responses

Billing and package tools return USD-first fields (for example pricePerMonthUsd, totalAmountUsd) plus exact ledger amounts as *Cents. Prefer *Usd when talking to users. Checkout price input is in dollars.

Tools

The hosted server exposes 70+ tools. Grouped by area:

Team and projects

Git and repository inspection

Applications and deployments

Container registry

Environment variables

Databases

Storage

Domains

Billing

Agent Plugin

The seenode Agent Plugin packages skills and client adapters for Cursor, Claude Code, and OpenAI / Codex on top of the same hosted MCP URL. Until marketplace plugins are published, install skills manually from that repo. See Agent Skills.

Related